This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
Google’s May 2026 core update finished yesterday. The dashboard entry closed with its standard line: “The rollout was complete as of June 2, 2026.” By Search Engine Land’s count it took about 12 days, inside the two weeks Google had allowed.
It was not a quiet one. Barry Schwartz at Search Engine Roundtable wrote that it felt larger than the March core update, with heavy volatility on the first weekend, again on Saturday, May 30, and another spike on the morning of June 2. His post on that last spike was updated to note that Google declared the update done about an hour after it went up. Volatility trackers are third-party measurements, not Google data, but the pattern was widely reported.
For reputation teams, this update makes a useful case because of what happened two days before it started.
Two changes, one window
On May 19 at I/O, Google made Gemini 3.5 Flash the default model in AI Mode for everyone. On May 21 the core update began. Anyone who saw the AI answer about their company change in late May has two candidate explanations, and they call for different responses.
If the core update caused it, the change came through ranking: different pages moved into the positions AI answers draw from. If the model change caused it, the sources may be the same, and the difference lies in how they were read and phrased.
Google does not label which system produced a given answer, and dates alone cannot separate two changes that overlap. There is still a practical way to get most of the way there.
Sources or sentences
Compare the cited sources, not only the text. For each query that matters, such as the company name, an executive’s name, or a product with a known controversy, look at the links supporting the AI Overview or AI Mode answer and at the classic results beneath it.
If the cited sources changed, and the classic results moved in the same direction, the core update is the more likely driver. The response is the familiar one: work out which pages gained and lost visibility, and whether the company’s own pages are among them.
If the sources are largely the same but the answer reads differently, leading with a different fact, taking a different tone, or including a detail it used to leave out, the model is the more likely driver. That calls for different work. It is less about ranking and more about whether the company’s key facts are stated plainly enough, on pages already being retrieved, that a different model reads them the same way.
When both changed, the honest conclusion is that they cannot be fully separated. That is still worth knowing, because it keeps a team from spending a month on the wrong fix.
Why one check is not enough
Google’s guidance on core updates recommends waiting at least a full week after a rollout completes, then comparing with a week before it began. That applies here too. The June 2 spike means the final state had little time to settle before the update was declared done. Sampling the same queries daily for a week, and keeping the source lists, gives a far better basis for diagnosis than a single check this morning.
I have argued before that provenance is becoming the product in generative search. This is the working version of that idea. When an answer about your company changes, the first question is not what it now says. It is what it is now reading.